This work presents a vision-based staircase identification system which comprises two sequential tasks: 2D staircase detection followed by 3D staircase localization. The 2D detector pre-screens the input image and the 3D localization algorithm continues the task of retrieving geometry of the staircase on the reported region in the image. The 2D staircase detector defies conventional thinking by exploring the issue from the machine learning perspective. It implements a binary classifier based on Viola-Jones rapid object detection framework. This thesis introduces a novel set of PCA-based Haar-like features which extends the classical Haar-like features from local to global domain and are extremely efficient at rejecting non-object regions at the early stages of the cascade. To adapt the framework to the context of staircase detection, modifications have been made on the scanning scheme, multiple detections integrating scheme and the final detection evaluation metrics. The “V-disparity” concept, originally used for ground plane estimation in autonomous navigation, is applied to detect the planar regions on the staircase surface. “V-disparity” can locate 3D planes quickly from disparity maps. However, its inherent drawbacks impede it from being used in wider applications. We have made improvements to the original “V-disparity” concept to adapt it to the application of 3D staircase localization.